|
Cost-effective semiconductor device manufacturing requires accurate process data and data analytical techniques that can quickly identify process deviations and root causes, in order to be able to suggest remedies. Such advanced data analysis methods accept process data from a variety of sources and time correlate it with production information, creating data sets that represent conditions at a specific production step, tool, chamber and process combination. The data is then used for control and process optimization purposes. There is a very strong interest in the implementation of process sensors such as RGA, V/I probe, HERCULES (SEERS), FTIR, Particle Counters and OES to provide such data sets for Advanced Process Control in state-of-the-art process tools. In this report, we describe the application to plasma and vacuum processes of Multivariate Analysis (MVA) Tools and Models. We discuss the issues that surround the implementation of this technology, including: the use of different and in-compatible data protocols, incomplete data, poor data quality, the need to analyze excessive amounts of data, the need to combine data from different sources and the lack of correlation found between vacuum and plasma process and quality issues. A live demonstration of real time FDC (Fault Detection), based on "SIMCA GEN 2 Batch Analysis" MVA together with PLS (for virtual metrology) will be presented. A Chamber Matching example will be used to illustrate the detection of differences between two plasma process chambers. |